Moonshot AI, the Beijing-based startup behind the Kimi model family, released Kimi WebBridge on May 14, 2026, a browser extension that gives AI agents the ability to interact with websites the way a human would, clicking buttons, filling out forms, scrolling pages, and extracting data, all while running entirely on the user’s local machine. The extension is available on the Chrome Web Store and the official Kimi website.
The distinction matters. Most AI browser automation tools route your browsing activity through cloud infrastructure, meaning the agent sees your logged-in sessions, your private page content, and whatever sensitive data happens to be on screen. Kimi WebBridge takes a different approach: it pairs a local background service with the browser extension, and the two communicate via Chrome DevTools Protocol, the same low-level interface developers use for debugging. Nothing leaves the device. Moonshot never sees the content.
What the Extension Actually Does
Kimi WebBridge works inside the Chrome or Edge window a user already has open, with all existing cookies and logins intact. It is not a sandboxed browser or a separate environment. An agent using WebBridge can open pages, click interactive elements, read text, take screenshots to understand what it is looking at, and pass results back to whatever AI tool is driving it.
The practical applications are straightforward: scan LinkedIn job listings across multiple searches and compile them into a spreadsheet; check the same product price across ten retailers and return the best deal; browse Amazon for a mechanical keyboard under $150 with at least 4.5 stars and produce a ranked comparison. Any task that involves repetitive clicking through a website, the kind that takes twenty minutes of manual work, becomes a single-sentence prompt.
Kimi WebBridge officially supports Kimi Code CLI, Claude Code, Cursor, Codex, and Hermes. It is not locked to the Kimi ecosystem: users of other AI coding or agent tools can connect WebBridge to their existing workflow via a connection command provided on the setup page. The underlying architecture supports up to 300 parallel sub-agents executing across 4,000 coordinated steps simultaneously, a scale that makes complex, multi-stage browser tasks tractable.
The Model Behind It: Kimi K2.6
The extension is powered by Kimi K2.6, Moonshot’s latest model, released in April 2026. K2.6 is a one-trillion-parameter mixture-of-experts model that now scores 58.6% on SWE-Bench Pro, a benchmark measuring real-world software engineering performance on actual GitHub issues. That puts it ahead of GPT-5.4 at 57.7% and Claude Opus 4.6 at 53.4%.
The original Kimi K2, launched in July 2025, ranked first among open-source models and fifth overall on the LMSYS Arena leaderboard. K2.6 extends that trajectory, and the WebBridge release is the first major product built explicitly around the model’s agentic capabilities rather than its raw benchmark performance.
(Related: Kimi K2.6 Becomes World’s Top Open-Weights AI Model, Closing to Within Three Points of GPT-5)
For Western users encountering Kimi for the first time in 2026, the model’s reputation may have arrived via an unexpected route: the Cursor controversy.
The Cursor Controversy
On March 19, 2026, the AI coding startup Cursor, valued at $50 billion, launched its Composer 2 model, marketing it as “frontier-level proprietary coding intelligence” built through “continued pretraining” and reinforcement learning. The announcement lasted less than twenty-four hours before a developer named Fynn intercepted API traffic and found a model identifier: kimi-k2p5-rl-0317-s515-fast.
Elon Musk posted three words: “Yeah, it’s Kimi 2.5.” Moonshot’s head of pretraining ran a tokenizer analysis and confirmed an identical match. Cursor’s VP of Developer Education, Lee Robinson, acknowledged the open-source base within hours, arguing that roughly 75% of the compute had gone into Cursor’s own training pipeline. Co-founder Aman Sanger called the omission “a miss from the start.” Moonshot, for its part, took the high road, officially congratulating Cursor on the launch.
The episode illustrated something that has become increasingly clear in 2026: Chinese open-weights models are not just competitive on paper. They are being quietly embedded into products marketed as proprietary Western AI, a dynamic that has drawn scrutiny from the U.S. State Department and congressional investigators alike.
A Crowded but Differentiated Market
Browser automation for AI agents is not a new category. Anthropic’s computer use feature lets Claude interact with desktops. OpenAI’s Operator and ChatGPT Atlas perform similar tasks through hosted cloud services. Google has DeepMind-powered agent experiments, and Perplexity has its Comet Browser.
The differentiator Kimi WebBridge is betting on is the local-first architecture. Cloud-based browser agents are convenient, but they require routing browsing activity through a third party. For anything involving personal accounts, financial data, or a company’s internal tools, that is a meaningful privacy consideration — and one that enterprise users in particular are likely to weigh carefully.
Whether that bet pays off will depend on adoption. WebBridge is available now on the Chrome Web Store and through the official Kimi setup page, with Mac and Windows support. The extension requires Kimi Claw Desktop, a local app that runs the background service.
Moonshot’s Broader Trajectory
Kimi WebBridge arrives at a moment when Moonshot AI is navigating a more complex operating environment than it was a year ago. The company neared a $2 billion funding round earlier this year as its ARR doubled to $200 million, and it has been weighing a Hong Kong IPO following its $18 billion valuation. At the same time, it has faced data security scrutiny after a reported incident in which Kimi allegedly leaked a stranger’s resume to another user, a reputational risk the local-first architecture of WebBridge directly addresses, at least for the browser automation use case.
The Kimi K2.6 code preview that EastFrontier covered in April showed the model outperforming Claude Opus 4.5 at 76% lower cost. WebBridge is the product layer that turns that cost-performance advantage into something a non-developer can use directly, no API key required, no cloud dependency, no data leaving the machine.
(Related: China’s Agentic AI Boom: How OpenClaw Drove Token Consumption to 140 Trillion Per Day)